Yesterday’s Results

Summary

## No High/Medium confidence results for January 17

High Confidence

Medium Confidence

Recent Performance

Last 30 Days

Current Season

Performance Chart

Today’s Picks

High Confidence

Medium Confidence

All Picks

Historical Backtest

All-Time P&L

Season Summary

Strategy Stats

Methodology

The Model

Elastic Net Regression - A regularized linear model that predicts player points.

Features Used (12 total)

Feature Description
pts_rolling_5/10/20 Average points over last 5, 10, 20 games
min_rolling_5/10 Average minutes played
fg_pct_rolling Field goal percentage (10-game)
usage_proxy FGA per minute
is_home Home court indicator
rest_days Days since last game
season_avg_pts Season average
games_played Games this season
consistency Scoring variability

How It Works

  1. Calculate features from recent games
  2. Model predicts expected points
  3. Compare prediction to sportsbook line = “edge”
  4. Only bet UNDERS with 3+ point edge

Why Only Unders?

The model systematically underpredicts player scoring. From walk-forward backtest:

Strategy Win Rate ROI
All bets 48% -8%
Overs only 44% -15%
Unders only 55% +8%
Unders + 4pt edge 57% +12%

Since actual points > predicted points: - UNDERS win when line > actual (and our low prediction helps) - OVERS lose because we already underestimate

Betting Rules

Rule Requirement
Bet Type UNDERS only
Min Edge 3+ points
Max Odds -115 or better
HIGH conf 4+ point edge

Bottom line: Only bet UNDERS. Skip overs entirely. Be selective.

Strategy Definitions

The backtest compares eight betting strategies to understand model value and optimal thresholds:

Strategy Min Edge Max Odds Bet Type Description
TRUE BASELINE (No Model) 0 pts -300 All Bet everything without model selection. Should LOSE money due to vig (~-5% ROI). Proves the model adds value.
UNDER BASELINE (No Model) 0 pts -300 Unders Bet all unders without model. Shows base rate advantage of unders (~52% win rate).
OVERS BASELINE (No Model) 0 pts -300 Overs Bet all overs without model. Shows base rate disadvantage (~48% win rate, loses more).
MODEL LOW (2pt) 2 pts -115 Unders Model-selected unders with 2+ point edge. Entry-level model threshold.
MODEL MEDIUM (3pt) 3 pts -115 Unders Model-selected unders with 3+ point edge. Core profitable strategy.
MODEL HIGH (4pt) 4 pts -115 Unders Most selective - only 4+ point edge. Fewer bets but highest win rate.
ALL BOOKS (3pt) 3 pts -115 Unders Same as MEDIUM but bets at ALL qualifying sportsbooks. Shows multi-book value.
STAR UNDERS (2pt) 2 pts -115 Unders Only players with line >= 20 points. Targets star player bias (60%+ UNDER win rate).

Key Insights

  • TRUE BASELINE loses money (-5% ROI) - proves the model adds significant value
  • OVERS lose more than UNDERS - due to natural base rate bias (52% UNDER win rate)
  • Model strategies are profitable (+18% to +25% ROI) - edge selection works
  • Star players have highest base rate bias - 60%+ UNDER win rate for high-scoring players
  • Higher edge = higher win rate - but fewer betting opportunities

Generated 2026-01-18 11:25 ET